Perplexity AI has proposed an ambitious plan to rebuild TikTok in the U.S., aiming to create a more transparent, trustworthy, and efficient platform. This initiative addresses concerns about TikTok's current algorithm and infrastructure.

Key Points:

1. Transparent Algorithm: Perplexity plans to make TikTok's "For You" feed algorithm open-source, ensuring users understand content recommendations and eliminating risks of manipulation.

2. Enhanced AI Infrastructure: Leveraging Nvidia Dynamo, Perplexity aims to scale TikTok's recommender models and improve content delivery speed, offering a smoother user experience.

3. Combatting Misinformation: By integrating citation and community note features, Perplexity seeks to enhance trustworthiness, allowing users to cross-reference information in real-time while watching videos.

Conclusion
Perplexity AI's proposal highlights a shift towards transparency and user empowerment in social media platforms. If successful, this initiative could redefine TikTok's role as a neutral and trusted space for content discovery.

Growing capability of AI in performing tasks with respect to its length

A recent research paper introduces a new way to assess AI progress: by measuring how long an AI can reliably complete a task without assistance. This approach focuses on real-world effectiveness rather than isolated benchmarks, revealing significant advancements in AI capabilities.

Key Points:

1. New Benchmark: Task Length AI is evaluated based on task duration it can complete with a 50% success rate, measured in terms of how long it would take a human professional. Current top models can handle tasks equivalent to one hour of human effort.

2. Exponential Progress: Over the last six years, the time length AI can handle has doubled approximately every seven months. Earlier models handled 5–10 minute tasks; now they manage tasks lasting about an hour, demonstrating rapid real-world advancements.

3. Future Potential: Though AI models excel at benchmarks, they struggle with consistency on extended tasks. However, if trends continue, AI might soon handle week- or month-long projects reliably, significantly expanding its utility.

Conclusion
This research highlights the transition of AI from simply answering questions to performing sustained, practical work. While challenges like consistency remain, the pace of progress suggests that reliable, independent AI systems capable of managing real projects may emerge in just a few years. Exciting times lie ahead for AI!

EXAONE Deep - Reasoning model by LG research

EXAONE Deep, developed by LG AI Research, is a powerful reasoning AI model excelling in mathematical logic, scientific reasoning, and programming tasks. Its performance demonstrates remarkable efficiency and capability, even at smaller model sizes.

Key Points:

1. Exceptional Benchmark: Performance EXAONE Deep 32B surpassed competitors like DeepSeek’s R1 on challenging benchmarks (CSAT 2025: 94.5; AIME 2024: 90.0), matching the larger DeepSeek R1 model (671B) in AIME 2025 despite being significantly smaller.

2. Versatility: Across Model Sizes Smaller models (7.8B and 2.4B) showed strong results, excelling in MATH-500, AIME, GPQA Diamond (66.1), and LiveCodeBench (59.5), ranking first in specialized benchmarks for reasoning and coding.

3. A Paradigm Shift in AI Development: EXAONE Deep 32B’s success over larger models, like DeepSeek R1, reflects a broader shift towards creating smaller, more efficient, and cost-effective AI systems without compromising performance.

Conclusion
EXAONE Deep’s achievements emphasize the rapid evolution of AI, showcasing how smaller models can deliver exceptional results. This trend highlights not just improving intelligence but also an increased focus on resource efficiency, setting new benchmarks for the future of AI innovation.

Useful AI Tools

1. Aha - AI influencer marketing team

2. Arcade AI - Helps you create interactive demos.

3. Sider AI - Deep research agent.

Thankyou for reading.